What excellent AIs are there in China that could replace ChatGPT?Large Language Model?

In the process of learning AI big language models, I compiled study notes on 12 domestic big language models, including the name of the big language model, R&D team, access link, model list, and introduction.

Since the model content and price of each company team may vary in different periods, it is specially noted that the time recorded in this article isJune 5 – June 6, 2024.

Moonshot

Official Website

https://www.moonshot.cn/

Introduction

Language model used by Kimi smart assistant. Moonshot's text generation model (moonshot-v1) is trained to understand natural and written language, and it can generate text output based on input. The moonshot-v1 model can be used for a variety of tasks, including content or code generation, summarization, dialogue, creative writing, etc.

Model List

Currently, the models supported by moonshot are:

moonshot-v1-8k: It is a model with a length of 8k and is suitable for generating short texts.

moonshot-v1-32k: It is a 32k-length model suitable for generating long texts.

moonshot-v1-128k: It is a 128k-length model suitable for generating very long texts.

The difference between the above models lies in their maximum context length, which includes the input message and the generated output, and there is no difference in effect. This is mainly to facilitate users to choose the appropriate model.

Product Pricing

Text generation model Moonshot-v1

Here 1M = 1,000,000, and the price in the table represents the price per 1M tokens consumed.

The difference between the above models lies in their maximum context length, which includes the input message and the generated output, and there is no difference in effect.

Charging interface description

• Chat Completion API charges: charged based on actual input and output tokens consumption
• File-related interfaces (file content extraction/file storage) are free for a limited time

Development Team

Moonshot AI is a domestic general artificial intelligence startup founded in March 2023. The company's founder is Yang Zhilin, a post-90s generation who holds a bachelor's degree from the Department of Computer Science and Technology at Tsinghua University and a doctorate from the Language Technologies Institute at Carnegie Mellon University. Yang Zhilin has a deep academic background, has published papers with many Turing Award winners, and has worked at top global artificial intelligence institutions such as Facebook AI Research and Google Brain.

In addition to Yang Zhilin, other co-founders of Dark Side of the Moon include Zhou Xinyu and Wu Yuxin, who also graduated from Tsinghua University and have extensive work experience in the field of artificial intelligence. Zhou Xinyu joined Megvii after graduating from university to study algorithm mass production, while Wu Yuxin worked at Meta's Artificial Intelligence Research Institute.

The team of Darkside of the Moon Technology Co., Ltd. stands out for its youth, professionalism and rich experience. It is committed to developing To-C products and has launched Kimi AI, the world's first smart assistant product that supports the input of 200,000 Chinese characters. The team has shown its strengths and is comparable to other senior leaders.Large ModelStartups have a completely different attitude and temperament, and are committed to providing individuals with more convenient, more powerful, and more personalized inclusive products through AI.

Thousand Questions on Tongyi

Official Website

https://tongyi.aliyun.com/

Introduction

Tongyi Qianwen is a large language model independently developed by Alibaba Cloud. It is used to understand and analyze the natural language input by users and provide services and assistance to users in different fields and tasks. You can get the results that meet your expectations by providing the clearest and most detailed instructions possible.

Model List

Product Pricing

Development Team

Tongyi Qianwen's R&D team is affiliated with Alibaba Cloud and brings together top experts in natural language processing, deep learning, computer science and other fields. The team focuses on developing ultra-large-scale pre-trained language models, such as the Qwen series, which have advanced capabilities such as multi-round dialogue, copywriting, and logical reasoning. They not only have strong technical strength, promoting the open source of models from 32 billion to 110 billion parameters, but also demonstrated outstanding innovation capabilities and cutting-edge technology levels.

Zhipu GLM

Official Website

https://zhipuai.cn/

Introduction

GLM stands for General Language Model, which is a pre-trained language model based on autoregressive filling-in-the-blank. The ChatGLM series of models supports relatively complex natural language instructions and can solve difficult reasoning problems. The model is equipped with an easy-to-use API interface, allowing developers to easily integrate it into various applications. It is widely used in many fields such as intelligent customer service, virtual anchors, chatbots, etc.

Model List and Pricing

Development Team

The research and development team of Zhipu GLM model is a top team from the Knowledge Engineering Laboratory (KEG) of the Department of Computer Science at Tsinghua University, with CEO Zhang Peng as the core. Team members have rich technical backgrounds and professional knowledge, including Liu Debing, a disciple of Academician Gao Wen of the Chinese Academy of Engineering and chairman of Zhipu Huazhang, and Wang Shaolan, a Ph.D. in Tsinghua Innovation and president of the company. The team focuses on the research and development of large-scale pre-trained models, launched the GLM large model, and achieved remarkable results in the field of artificial intelligence. They are committed to applying advanced AI technology to practical scenarios and continuously promoting the company's technological innovation and application in government services, enterprise services and other fields.

Bean bag large model

Official Website

https://www.volcengine.com/product/doubao

Introduction

Doubao Big Model is a self-developed big model launched by ByteDance. It has been verified by ByteDance's internal 50+ business scenarios, and is continuously polished with a large daily usage of hundreds of billions of tokens. It provides multi-modal capabilities and creates a rich business experience for enterprises with high-quality model effects.

Model List and Pricing

Development Team

The Doubao Big Model Development Team is a team under ByteDance, established in 2023. The team is committed to developing the industry's most advanced AI big model technology, becoming a world-class research team, and contributing to the development of science and technology and society. The team has a long-term vision and determination in the field of AI. Its research directions cover NLP, CV, voice, etc. It has laboratories and research positions in China, Singapore, the United States and other places. Relying on the platform's sufficient data, computing and other resources, the team continues to invest in related fields and has launched a self-developed general big model that provides multimodal capabilities. The downstream supports more than 50 businesses such as Doubao, Button, and Jimeng, and is open to corporate customers through Volcano Engine.

Wenxin Large Model

Official Website

https://wenxin.baidu.com/

Introduction

The Wenxin Big Model is an industrial-grade knowledge-enhanced big model developed by Baidu. It is based on the PaddlePaddle deep learning platform and uses knowledge-enhanced technology to achieve deep understanding and generation capabilities for massive data. The Wenxin Big Model not only covers multiple fields such as natural language processing, vision, and cross-modality, but also greatly improves the model's zero-sample/small-sample learning capabilities through continuous learning and optimization.

Model List

Product Pricing

Development Team

The Wenxin Big Model was developed by the Baidu team, which focuses on the research and application of basic technologies for pre-trained big models, and has achieved remarkable results in the fields of NLP and multimodality. The team has launched the ERNIE series of models, including multiple versions from ERNIE 1.0 to 4.0, as well as multimodal big models such as Wenxin Yiyan (ERNIE-Bot) and ERNIE-M, and has topped many authoritative lists. The team members are composed of doctoral and master students in the fields of artificial intelligence, computer science, etc., with deep algorithm implementation capabilities and understanding of cutting-edge NLP technologies.

Hunyuan Large Model

Official Website

https://hunyuan.tencent.com/

Introduction

Tencent Hunyuan is a large language model developed by Tencent. It has strong Chinese creation capabilities, logical reasoning capabilities in complex contexts, and reliable task execution capabilities. Tencent Hunyuan has promoted performance improvement and reduced reasoning costs by adopting a mixed expert model (MoE) structure. It has excellent performance in Chinese performance, especially text generation, mathematical logic, and multi-round dialogue, and is at the industry leading level overall. Currently, it supports up to 256K contexts; in the long text input scenario, the needle-in-the-haystack index reaches 99.9%.

Model List

Product Pricing

Development Team

Hunyuan Big Model was developed by Tencent's Hunyuan Big Model team, which is composed of top AI scientists, engineers and researchers. It is committed to research and innovation in deep learning, natural language processing, computer vision and other fields, aiming to create a world-class intelligent model. The team members have rich academic background and practical experience, have published important papers in many top international conferences, and have won the championship of related competitions many times.

Tiangong Model

Official Website

https://model-platform.tiangong.cn/

Introduction

The Tiangong model is a dual-100 billion-level large language model developed by Kunlun Wanwei. The Tiangong model can meet diverse needs such as copywriting, knowledge question and answer, code programming, logical deduction, mathematical reasoning, etc. The model passed the first round of algorithm engineer interviews and demonstrated its ability to understand programming languages and natural languages including Chinese. The Tiangong model also has powerful natural language processing and intelligent interaction capabilities, and can realize personalized AI search, intelligent question and answer, chat interaction, text generation and other application scenarios.

Model List and Pricing

Development Team

The Tiangong AI development team was jointly developed by Kunlun Wanwei and Singularity Intelligence, a leading domestic AI team. Kunlun Wanwei was founded in 2008 and focuses on fields such as games, social media, and e-commerce. In 2023, Kunlun Wanwei and Singularity Intelligence officially reached a comprehensive technical cooperation and jointly developed the Tiangong AI assistant, which has powerful natural language processing and intelligent interaction capabilities. In addition, the Tiangong AI development team also includes some well-known scientific researchers, such as Professor Yan Shuicheng, who is an international top expert in computer vision and machine learning. He joined Kunlun Wanwei and served as co-CEO of Tiangong Intelligence with Kunlun Wanwei founder Zhou Yahui, and also served as the dean of Kunlun Wanwei 2050 Global Research Institute.

iFlytek Spark Large Model

Official Website

https://xinghuo.xfyun.cn/spark

Introduction

iFlytek Spark Big Model has cross-domain knowledge and language understanding capabilities, and can complete tasks such as question-answering dialogues and literary creation. It continuously learns and evolves from massive text data and large-scale grammatical knowledge, and realizes a closed loop of the entire process from raising questions, planning problems to solving problems.

Model List and Pricing

Development Team

iFlytek Spark Big Model was developed by an elite team of iFlytek that focuses on the research and development of artificial intelligence technology. The team consists of a group of experienced scientists, engineers and technicians who have deep professional knowledge and practical experience in the fields of artificial intelligence, deep learning, natural language processing, etc. The team members are committed to researching and developing advanced artificial intelligence technologies to promote the development of intelligent speech recognition, machine translation, intelligent dialogue systems and other fields.

Discussing large language models

Official Website

https://platform.sensenova.cn/

Introduction

SenseChat, the full name of Shangliang Big Model, is a Chinese language big model application platform developed by SenseTime. This model was first made public at SenseTime's Technology Exchange Day on April 10, 2023. The base model of Shangliang Big Model is "Shusheng Puyu InternLM-123B" released by SenseTime and several domestic scientific research institutions. It has 123 billion parameters and ranks second in the total test score of 300,000 questions in the world's well-known evaluation set.

Model List and Pricing

Development Team

The Shangliangda model was developed by SenseTime, a world-leading artificial intelligence company committed to advancing original technology research and continuously enhancing industry-leading multi-modal, multi-task general artificial intelligence capabilities, covering key technology areas such as perception intelligence, natural language processing, decision intelligence, and intelligent content generation, as well as key capabilities including AI chips, AI sensors, and AI computing infrastructure.

Yi-Large Model

Official Website

https://platform.lingyiwanwu.com/

Introduction

Yi-Large is a closed-source model with hundreds of billions of parameters developed by Zero One Everything Technology Co., Ltd., led by Kai-Fu Lee. The model has performed well in many evaluations, including ranking first in Win Rate on Stanford's AlpacaEval 2.0 global large model rankings, and ranking first in Chinese language ability in SuperCLUE's April benchmark performance. In addition, the Yi-Large model has also demonstrated strong performance in general capabilities, code generation, and instruction compliance, firmly ranking among the world's first-tier large models.

Model List and Pricing

Development Team

The Yi-Large language model was developed by the R&D team of 01.AI. The team is characterized by its diversity and professionalism. The team members come from all over the world with different cultural backgrounds and professional skills. They have deep experience and expertise in large model technology, natural language processing, machine learning, system architecture, data security and product innovation.

The team is led by Dr. Kai-Fu Lee, a pioneer and expert in the field of artificial intelligence with more than 40 years of industry experience. Under his leadership, the team is committed to promoting the development of artificial intelligence technology and sharing technological achievements with developers and researchers around the world through open source projects and collaborations.

MiniMax Large Model

Official Website

https://platform.minimaxi.com/

Introduction

MiniMax is a leading domestic startup that has a language model with hundreds of billions of parameters that integrates text, voice, and vision, and has opened up the entire product chain. By building a super-large-scale experimental platform and reasoning platform, running through the iterative closed loop of technology and products, it has achieved rapid growth in model capabilities and user interaction.

Model List and Pricing

Development Team

Founded in December 2021, MiniMax is a leading general artificial intelligence technology company committed to co-creating intelligence with users. MiniMax's development team is composed of a group of experts who are passionate about artificial intelligence, with rich industry experience and deep technical background. Team members include data scientists, software engineers, linguists, and machine learning experts, who work together to optimize and innovate models.

Step Large Model

Official Website

https://www.stepfun.com/

Introduction

Currently, Step-1 and Step-2 preview versions have been launched by Step-Star. They are the classic large model step-1 with hundreds of billions of parameters and the large model MoE architecture with trillions of parameters. They can handle various complex language tasks. Whether it is text creation, cross-language communication, or problem solving, logical reasoning, etc., it can improve the work efficiency of users. In addition, this series of models also has excellent mathematical and coding capabilities, and can also play an important role in scientific computing, program development and other fields.

Model List and Pricing

Pricing has not been revealed.

Development Team

Step Star is a startup company focused on the research and development of general large models. At the Global Developer Pioneer Conference in March 2024, they released the Step series of general large models for the first time, including the Step-1 100 billion parameter language large model, the Step-1V 100 billion parameter multimodal large model, and the Step-2 trillion parameter MoE language large model preview. The team consists of about 150 people, and the core members include founder and CEO Jiang Daxin, system director Zhu Yibo, and data director Jiao Binxing. They have rich research and practical experience in the field of AI and are committed to promoting the development of AI technology.